Field notes for imagined worlds.
Practical research on AI image and video tools, motion, prompting, and the systems that turn experiments into finished work.
- field notes
- 65
- directions
- 05
- archive
- 2026
65 results
ByteDance World Model, Krea 3, Sora and OmniHuman 2.0
The most important story is not one model winning. It is the creative stack becoming editable at every layer: performances from audio, images changed without rebuilding the frame, navigable worlds with authored camera paths, and open post-production models for mattes, HDR and high-resolution finishing. This guide separates products you can use now from demonstrations, reported negotiations and sponsored claims.
Adobe vs ArtCraft: PhotoCraft, FilmCraft and EffectCraft Tested
Can free, open-source creative apps built with AI replace Adobe today? AI Samson gives both sides the same files and briefs, then measures output quality, completion time and hands-on usability. The results are more useful than the headline: PhotoCraft wins several simple automation races, FilmCraft handles structured edits but exposes major workflow gaps, and EffectCraft delivers the strongest surprise in motion graphics.
Opus 5.5 Effort Test: Six AI Music Videos Compared
One song, one deliberately open brief and six autonomous Claude Code runs. Devarf's experiment shows what higher Opus 5.5 effort can improve, why a cheap 2D result can still work, and how Ultracode can turn a creative test into a runaway multi-agent job. This guide reconstructs the setup, includes the exact prompt, separates measured facts from subjective quality and adds a safer benchmark workflow.
ArtCraft Hands-On: Seven Free Adobe Alternatives Tested
AI Samson installs ArtCraft's familiar creative apps, compares them with Adobe workflows and lets Claude edit native projects through MCP. The impressive part is not visual imitation. It is the combination of editable files, direct agent control and human correction. This guide organizes every test, separates creator observations from official project claims and turns the demo into a safe evaluation plan.
Claude Motion Tutorial: Charts, Explainers and Video Sync
Paul J Lipsky turns a 90-day recovery spreadsheet into an animated story, rebuilds an AI-agent explainer with visual references, aligns graphics to a YouTube intro with an SRT transcript and adds motion inside a Claude Dashboard. This guide organizes the complete workflow, the prompts and the checks that keep the result accurate.
ArtCraft Review: Seven Free Adobe Alternatives Explained
Rough Cut Pete asks whether Adobe can survive after Brandon Thomas and the ArtCraft community released seven familiar creative apps in roughly a week. The code is real and the pace is extraordinary, but repository creation is not feature parity. Here is what is available, what independent testing found, what the licenses permit and how to evaluate the alpha safely.
Claude Motion, Dashboards and Haiku 5.5 Explained
Brock Mesarich covers a crowded week of AI releases. This guide separates shipped features from creator tests and forecasts, preserves his audience-retention Motion prompt, and adds a skeptical seven-day service pilot that keeps publishing, purchases, private messages and deployment behind human approval.
Perchance AI Image Generator Guide and 5 Prompts
HiddenByte tests Perchance with five visual ideas, from a futuristic African city to an alien expedition. This guide turns that walkthrough into a repeatable workflow, adds five polished prompts you can paste and customize, and separates verified Perchance features from settings that may change over time.
Claude JavaScript Video With Canvas and FFmpeg
Chase AI shows how Claude can write a deterministic animation as a function of time, capture every frame in a headless browser and encode the result locally. This guide preserves his Level 1 and Level 2 prompts, explains the self-review loop, verifies the public Animate repository and compares custom Canvas with Remotion and HyperFrames for real production work.
How Jason Lee Builds Motion Graphics With Opus 5.5
Jason Lee starts with a code-built QuickSend launch film, then adds an AI actor, a locked office, multi-angle start frames and Seedance 2.5 footage. This guide reorganizes his video and public Motion Graphics with Opus 5.5 playbook into one practical production system, including Motion OS, model settings, credit estimates, reusable prompts and the failure modes worth planning around.
How to Make Viral Motion Graphics With Claude for $0
Jad M.H tested the viral Opus 5.5 motion-graphics format at three levels: a simple silent reel, a lyric film with locally generated music and sound, then a directed short story with 375 effects and a self-critique loop. This guide explains what changed, what the '$0' claim really means and how to turn the process into a reusable Claude skill.
Build a Motion Design Studio With Claude Opus 5.5
Movez (@0xMovez) mapped the first wave of Opus 5.5 code-generated motion films into a studio pipeline. This guide turns that research into a practical build system: references, brand assets, a seek(t) renderer, springs, beat-aware audio, contact-sheet review, multi-format export and a reusable motion-reel skill.
AI Paper Animation With OpenArt: GPT Image 2 and Seedance 2.5 Prompts
Thomas Creates turns one visual grammar into explainers, short stories, product ads, seamless transition chains and a jointed paper puppet. This practical guide explains why the system stays consistent, how to build each reference asset and how to use every prompt from the Paper Animation Playbook.
ChatCUT in ChatGPT: AI Video Editor Workflow and Pricing
Marcin AI turns a raw talking-head recording into a tightly cut, motion-heavy intro by directing ChatCUT from the ChatGPT desktop app. This guide reconstructs the workflow, explains what is truly happening behind the interface, organizes reusable prompts and adds the pricing context the headline leaves out.
Claude Opus 5.5 Code-Only Motion Design Workflow
zero (@twoclipping) published a complete system for turning a reference, product story and beat-driven direction into polished motion design without After Effects. This guide reorganizes the method into a repeatable production pipeline, includes the four original examples and adds reusable prompts for planning, rendering and quality control.
Kling 4, Gemini 4 Argon, HeyGen Video and Eleven v4
Theoretically Media tests Kling 4.0 Flash across text-to-video, image-to-video and Omni scenes, then checks the rest of a crowded release week. The useful question is not which announcement sounds biggest. It is what creators can access today, what each model actually costs, and which claims still need independent testing.
Seedance 2.5 Cinematic AI Video Tutorial with Claude
Rourke Heath's one-hour tutorial treats AI filmmaking like production, not prompt roulette. The process begins with a visual language and a story, turns characters, locations, props, camera choices and performance notes into reusable references, generates longer Seedance sequences for continuity, then mines the best moments into a finished edit.
Midjourney September 2026 Update: Edit, Tile, V9 and SREF
Woollyfern's September roundup separates the changes you can use now from the roadmap still taking shape. Targeted edits preserve more of the source image, repeated edits degrade less, V8.1 and V8.2 tiles lose their faint seams, and the alpha website can preview one prompt across saved and featured SREF styles. Camera control, V9, Niji 8, new upscalers and the next video model remain previews rather than finished releases.
30 Claude Opus 5.5 AI Video Use Cases and Prompts
AI Samson's collection shows a broader shift than automated cutting. Opus 5.5 can direct video models, build editable browser animation, operate creative software and turn repeatable production rules into tools. This guide organizes all 30 creator examples, preserves every original X source and includes the reusable prompts Samson published.
Claude Opus 5.5 AI Video Workflow with Artlist MCP
Tao Prompts separates generative footage from precision graphics. Artlist MCP produces the source images and Seedance 2.5 clips inside Claude, then Opus 5.5 analyzes the clean plates and builds the readable labels, HUD elements, data overlays and synchronized sound design as a controlled second layer.
Claude Opus 5.5 Video Editing Workflow with Borumi
Paul J. Lipsky connects Claude Code to Borumi through MCP, gives the model a repeatable cleanup brief, then directs motion graphics and missing screen captures in plain English. The result gets him about 80% of the way to a finished YouTube tutorial, with the final creative pass still firmly human.
Claude Opus 5.5 AI Shorts Workflow: Shortify + 11 Prompts
Duncan Rogoff turns one topic into a polished vertical video with an AI avatar, motion graphics, captions, music and a comment-to-DM ending. This guide maps the full Shortify workflow, checks the creator-reported $8.34 cost, and organizes all 11 prompts from the companion PDF.
ChatGPT Images 2.5: 5 Creator Workflows and OpenArt Prompts
One reference image can help keep a character recognizable across scenes. This OpenArt tutorial also covers focused edits, infographics, channel graphics and motion design built from a verified still with Seedance 2.5. Explore the five workflows and seven original PDF prompts, with practical checks for text, facts and continuity.
AI Film Workflow: Codex, Runway MCP, and DaVinci Resolve
Tim Simmons remade his film The Bridge in about a day and a half. GPT-6 Astra in Codex helped rebuild the script and shot list, Runway MCP coordinated Seedance 2.5 generations, and DaVinci Resolve handled color, titles and sound. Tim still assembled the first cut himself. Watch the original, the remake and the making-of, then follow the workflow behind them.
ComfyUI Desktop: Guide to Generating AI Images Locally
ComfyUI looks like a wall of boxes and wires, but the first image doesn't require designing a workflow from scratch. In this GeekyDK tutorial, we start with the official installation, load the Z-Image Turbo template, resolve the missing models and understand what each part of the pipeline does. We also separate the real benefits of local processing from the exaggerated promises about cost, privacy, and consistency.
GPT-6 Astra and /animate: edit video with a refined skill
A recording of eight scripts became a series of vertical videos with cuts, graphics and 3D assets. Jay E's demonstration doesn't prove that just writing any prompt is enough: it shows how to create an initial skill, evaluate its defects and save the corrections for the next work. The attached PDF explains another skill, /generate, that can provide assets to the same system.
Midjourney Edit V8.2: Quick Edit test, references and Alpha
The new Edit model impressed Future Tech Pilot with its speed and natural-language controls. The same test also showed plastic-looking skin, unstable anatomy, limited variation and a style selector that barely changed the image. This guide separates confirmed capabilities from the creator's observations and offers a methodical test workflow.
Seedance 2.5: 33 animation styles and organized prompts
Jack, from the Atomic Gains channel, tests dozens of animation languages, from 1930s cels to retro anime, ink, stop motion, live action hybrids and CRT visuals. This article organizes the 33 master prompts from the video's accompanying guide into easy-to-explore families, with instructions for adapting them to your own story.
GPT-6 Astra and Higgsfield: 30 examples and 7 prompts
Atomic Gains tests GPT-6 Astra as a creative orchestrator and connects it to Higgsfield to generate assets, build websites, create gesture games, edit video, produce motion graphics and transform references into editable 3D scenes. This guide organizes the 30 examples, distinguishes demos from official capabilities, and publishes, with credit, the seven full prompts found in the accompanying document.
Midjourney Edit: August 2026 Model Guide
The August update turns editing in Midjourney into an instruction-driven workflow. Woollyfern demonstrates five ways to enter the new Edit model, clothing and style changes, consistent characters, inpainting, outpainting and reference combinations. This guide organizes the process, distinguishes official documentation from first impressions, and updates the status of the Alpha interface and models still in development.
Artlist MCP on Claude and ChatGPT: AI-powered guide and trailer
Artlist demonstrates how to connect image, video, voice and music generation to a conversation with Claude or ChatGPT. The underwater trailer shows the strength of the system, but also reveals the reality of production: planning, generating and iterating within the chat, using the Artlist interface when the reference requires more control and finishing editing, color and sound in an editor. This guide organizes the complete workflow, current limits, and reusable prompts with approval before spending credits.
GPT-Image 2.5: tests, Astra, Flare and Sunburst
Theoretically Media subjected GPT-Image 2.5 to tests that expose both the model's visual intelligence and its limits: a clock showing 5:15, editions with the passage of time, mirrors, characters in four angles, narrative compositions and cinema scenes. This guide organizes the results, separates observation from official specification and transforms the best ideas into a reproducible working system.
GPT Image 2 vs 2.5: Flare and Sunburst in 15 tests
Brendan Jowett placed GPT Image 2, GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst side by side in fifteen orders: poster, photography, product, technical diagram, multilingual lettering, character, architecture, physics and a surreal composition. This article organizes the results, corrects the limitations of the protocol and turns the comparison into a practical choice between speed, accuracy and total approval cost.
ChatGPT Images 2.5: Sketch, edits and templates
Paul J Lipsky explores ChatGPT Images 2.5 as a conversational creation tool: draw a room, fix elements with comments, remove backgrounds, restore a childhood photo, redesign a garden, turn your dogs into an illustration, and build a thumbnail by iteration. The results show where new experience accelerates ideas and where spatial selection, templates and human validation remain indispensable.
ChatGPT Images 2.5: Sketch, templates, Sunburst and Flare
OpenAI launched ChatGPT Images 2.5 with faster generation, greater fidelity to references, localized edits and new tools to go from an idea to the final result. This guide organizes the official announcement, incorporates demos of Images 2.5, Sketch, templates and API, explains when to choose Flare or Sunburst and converts the news into a practical workflow for creators, designers and product teams.
ChatGPT Image 2.5: testing infographics, consistency and PNG
Futurepedia put ChatGPT Image 2.5 in front of problems that interest real producers: posters full of information, localized changes, recurring characters, transparent cropping, style transfer and photorealistic scenes. This guide organizes the results, distinguishes the creator's tests from OpenAI's documented capabilities, and turns the findings into a verifiable workflow with practical prompts.
GPT-6 Astra and Images 2.5: 50 use cases with prompts
AI Samson brought together 50 experiences from the first week of GPT-6 Astra and ChatGPT Images 2.5, between robotics, real-time vision, Blender, games, design, video and science. This guide maintains companion order, credits each creator, distinguishes demonstration from proof, and adds a usable prompt for each case. At the end, there are also the ten complete prompts that the creator proposes to try without special hardware.
invideo Agent Two: film with AI from briefing to timeline
Vaibhav Sisinty created a visual-comedy short in which a man balances a car on one finger, rides through a park and flies over the city on a dragon. None of those shots were filmed. More useful than the spectacle is the persistent workspace that keeps the Director's Bible, cast, locations, rules, generations and edit together. This guide turns his invideo Agent Two demo into a repeatable workflow and explains why natural-language direction still needs human judgment.
H3 MAX in real time: AI TV, gaming and video on the go
The H3 MAX, a post-trained version of the MiniMax H3 created by fal, can generate a five-second audio clip in less than three seconds according to tests published by the company itself. This latency reduction already powers generative TV channels, chat-driven experiences, and games where the next shot is produced while the current one is still playing. This guide separates benchmark from real experience, open H3 from hosted H3 MAX, rapid generation from continuous streaming, and turns Theoretically Media video into a practical plan for experimenting with this new language without losing control of cost, continuity, security, or rights.
ComfyUI MCP: Claude and Codex create local workflows
The official ComfyUI MCP allows Claude, Codex, and other agents to inspect the local installation, find models and nodes, build workflows, validate graphs, and perform natural language generations. The learning curve decreases, but hardware, cost, privacy, creative control and review decisions do not disappear. This guide turns Theoretically Media testing into a practical system for working locally, turning to the cloud only when it's worth it, and keeping the human responsible for each run.
Seedance 2.5: 50+ examples, limits and practical prompts
AI Samson walks through more than 50 experiments published in the first week of Seedance 2.5, from lip sync rap and product announcements to clay render control, extensive casts and video transformation. This guide organizes examples by capability, separates demonstration from commercial promise, exposes physics and continuity flaws, and converts research into a production method with reusable prompts.
VFX vs Seedance 2.5: will AI change cinema?
ErikDoesVFX films a dragon sequence in Iceland and produces two versions: one with 3D assets, animation, Houdini simulations, rendering and compositing; another with Seedance 2.5 on Higgsfield. The first better preserves vision and continuity. The second works quickly and surprises when it can reinvent everything. This guide breaks down the comparison, shows the hidden costs on both sides and proposes a hybrid workflow to decide shot by shot.
How to create cartoon animations with AI: complete guide
Dan Kieft turns photographs into character sheets, builds episode frames in OpenArt, and animates scenes with dialogue in Seedance. This guide reorganizes the demo into a repeatable production system, explains the two vocal consistency methods, compares start/end frames with video references, and includes an original prompt package to create your own cartoon aesthetic.
Seedance 2.5 Long Video: 3-minute film test
JSFILMZ uses Dreamina's new Long Video beta to generate the first three minutes of an 18-minute short in a single request. The test combines four character sheets, a translated prompt compressed to 3,721 characters, 20,000 credits, and a 25- to 30-minute render at 480p. The result maintains the cast, dialogue and several narrative beats, but changes the geography of the diner, simplifies actions and loses important instructions. This guide explains what actually happened and offers a safer prompt structure for repeating the experience.
How to create a car ad with AI: complete workflow
Adil from Higgsfield AI rebuilds a cinematic car ad without a camera, crew or studio. The result comes from a system much more useful than a single generation: character sheets, cleaned locations, references saved as Elements, a skill from Claude to write Seedance 2.0 prompts and dozens of editorial decisions. This guide organizes the complete process, explains why errors happened, and offers a unique pack of adaptable prompts for each asset and scene.
MiniMax H3 image-to-video: practical test on fal.ai
Brian from the Creative AI Show put the MiniMax H3, also touted as the Hailuo 3.0, through its paces with portraits, a client chart, science fiction, a group image, and martial arts movement. The results show convincing emotion, detail and body movement, but also reveal a clear lack of adherence to the prompt and instability in the background. This guide separates the five tests specific to the promotional example, confirms the specifications, compares the price of fal.ai with the direct API, and proposes a more rigorous benchmark.
MiniMax H3: 2K video, native audio and references
The HitPaw Edimakor channel presents MiniMax H3 as a model capable of understanding text, image, video and audio within the same creative context. The proposal is real, but the less than three-minute demo omits important differences between the 2K API, the open base model, the territorial license and the integration announced for the publisher. This guide confirms specs, calculates costs, and turns launch into a testing workflow for filmmakers, brands, and creative teams.
Fable 5 for AI Filmmaking: 3 Free Tools
Dan Kieft transformed three repetitive production tasks into a set of applications: one organizes story, characters and locations; another blocks the camera and poses in a 3D scenario; the third offers 46 camera movements with reusable prompts. This guide explains the workflow, confirms what's actually free, details costs and API key security, and shows how to create your own tool without confusing a convincing prototype with a finished product.
Seedance 2.5: review, 30 seconds and 50 references
AI Samson brings together a first wave of Seedance 2.5 tests and shows why the update is of interest to directors: 30-second sequences, up to 50 multimodal references, more expressive acting, complex camera movements, storyboards, green screen and 3D preview. This guide confirms what the official documentation supports, separates curated examples from benchmarks, and turns the new features into a verifiable production workflow.
Higgsfield terms: training, outputs and privacy
Theoretically Media's Tim sparked an important discussion by analyzing terms submitted by Higgsfield on July 23, 2026. Three days later, the company published a review. This guide compares the two official texts, confirms the improvements, identifies the remaining training consent, and turns the debate into a checklist for creators and agencies.
Claude Skill vs Higgsfield: fal.ai, WaveSpeed and Kie.ai
Jay E of RoboNuggets demonstrates a /generate skill that routes Claude's image and video requests across models and providers, with budget controls, fallbacks, prompt logs and a local gallery. This guide examines Higgsfield's revised terms, compares subscriptions with pay-as-you-go APIs and explains what a safer alternative needs.
Automate Videos Vox Style with Claude and Higgsfield MCP
Zubair Trabzada shows how a detailed prompt, or a reusable skill file, can coordinate research, scripting, narration, images, graphics, sound and editing for a 60-second explainer. This guide reconstructs the method, checks what Claude and Higgsfield support, explains the costs and shows where human review remains essential.
FLUX 3 Video Test and Gossip Goblin Theatrical Release
Theoretically Media tests Black Forest Labs' new multimodal video model across trailer generation, image-to-video, and a 15-shot coherence challenge, then studies the human-led workflow behind Gossip Goblin. Here is what FLUX 3 officially supports, what remains an early-access observation, why Pomegranate is not the theatrical feature, and what filmmakers can borrow from Zack London's modular production method.
Truepix AI Reverse Engineer: Video Ads Into Templates
Truepix AI demonstrates a compact workflow for turning a reference video into a shot-by-shot blueprint with editable variables, reference images, dialogue, camera direction, and Auto Fill. This practical guide explains what the three-minute demo proves, what still needs testing, and how to adapt creative structure without copying a competitor's identity or claims.
Seedance vs Grok vs Kling vs Gemini Omni: Blind Test
Dan Kieft compares four leading AI video routes across interaction, motion transfer, dialogue, complex prompts, reference control, cinematic action, speed, and cost. This independent guide maps every result, discloses the Higgsfield sponsorship, corrects details that changed after publication, and turns the test into a practical model-routing workflow.
Seedance 2.5 Preview and Seedance 2.0 4K Test
Dan Kieft separates an ambitious Seedance 2.5 preview from a hands-on Seedance 2.0 4K test in Higgsfield. This evidence-led guide maps the 30-second and 50-reference claims, explains what remains unverified, compares the recorded resolution costs, and turns the test into a practical AI filmmaking workflow.
Two Open-Source Remotion Brand Videos: Magnific and Higgsfield
Naveen Annam has released two complete code-driven brand-video recreations through Creativly.ai. This practical guide explains what is inside the Magnific and Higgsfield repositories, how their Remotion architecture works, what to study first, and how to adapt the projects without carrying over another brand's identity.
Advanced AI Filmmaking: Character and Scene Continuity
Dan Kieft reverse-engineers his prehistoric AI short from pre-production through the final edit. The practical lesson is that cinematic continuity comes from reference systems, controlled performances, selective regeneration, and editorial judgment, not one perfect prompt.
How Kimi K3 Builds Cinematic Scroll-Driven Websites
Nick Saraev turns an AI-generated macro journey into a scroll-scrubbed landing page, then uses Kimi Code to scaffold the experience and publish it. This breakdown separates the creative asset, coding workflow, real cost, performance work, and production safeguards.
How to Set Up Palmier and Connect Claude or Codex
Palmier co-founder Marcos demonstrates the first-run workflow for Palmier Pro: install the local-first Mac editor, connect an AI coding agent through MCP, let the agent inspect and rough-cut a transcript, then add generated media without leaving the timeline.
How to Mix AI With Real Footage for Cinematic VFX
AI Samson demonstrates a hybrid AI VFX workflow that keeps real performance and camera movement as the foundation, then uses Gemini Omni Flash, reference images, GPT Image 2, and structured prompts to transform the shot.
How to Switch Styles in Midjourney with Vary, Remix, and SREF
Woollyfern demonstrates a practical Midjourney workflow for restyling one strong image with Remix Subtle and SREF codes, then building larger environment and outfit changes in controlled rounds.
Remotion Masterclass: The Advanced AI Motion Graphics System
FRMWRKD-EXPLAINED and Naveen Annam unpack a repeatable system for turning reference brand videos into reusable Remotion scenes with Gemini, Claude Code, contact sheets, and an iterative grading loop.
How to Edit a Viral Talking-Head Using ONLY AI (2026 Style)
A practical breakdown of Joseph's six-step talking-head workflow, from a timestamped podcast transcript to captions, branded AI animation, and a final Premiere Pro assembly.
Coming Back to Midjourney? 18 Updates You Missed in the Last 12 Months
A fast, practical guide for returning Midjourney users: what changed, what matters, and what to try first if you want better images, cleaner projects, and a faster workflow.